Sleep ECG signals reportedly add to 10-year heart-failure risk prediction

Deep learning on a single ECG lead during sleep

ScienceAlert reports that researchers trained a deep-learning system on sleep-study ECG data from nearly 16,000 patients and tested it on two further datasets of nearly 10,000 and 13,000 patients.

The reported model used electrical activity from a single chest ECG lead during sleep, together with sleep stages, to sort participants by long-term risk of atrial fibrillation, stroke, heart failure, and death from any cause. The article says the findings were published in SLEEP.

The report does not provide a directly comparable before-and-after clinical outcome or evidence that this approach is in clinical use.

According to the report, adding the sleep ECG-based deep-learning score improved prediction beyond established risk factors and sleep characteristics for several 10-year cardiovascular outcomes. It describes potential future screening use, not a deployed clinical tool or demonstrated patient benefit.

Improved prediction and clinical benefit need separate tests

The reported result suggests that ECG information already captured during sleep studies may contain predictive signals not fully represented by conventional cardiovascular risk factors. Whether incorporating it into care improves outcomes remains an identified next step.